TLDR: A comprehensive report commissioned by the European Parliament’s Committee on Legal Affairs highlights a critical mismatch between current EU copyright law and the rapid advancements in generative artificial intelligence (AI). Published in July 2025, the study, ‘Generative AI and Copyright: Training, Creation, Regulation,’ identifies five key findings, emphasizing that existing Text and Data Mining (TDM) exceptions are ill-suited for AI training, fully machine-generated outputs should remain unprotected, and a statutory remuneration scheme is essential for creators. The report calls for clear rules on input/output distinctions, harmonized ‘opt-in’ mechanisms, enhanced transparency, and equitable licensing models to safeguard creativity and ensure fair compensation in the evolving digital landscape.
The European Union is grappling with the profound implications of generative artificial intelligence (AI) on its copyright framework, as detailed in a pivotal report commissioned by the European Parliament’s Committee on Legal Affairs (JURI). The study, titled ‘Generative AI and Copyright: Training, Creation, Regulation,’ published in July 2025, underscores the urgent need for legislative reform to address the structural risks posed by AI to Europe’s creative economy. This comes as the EU’s landmark AI Act, while introducing transparency, is seen by many as falling short in protecting creators’ rights.
The Core Challenge: A Mismatch in Copyright Law
The report identifies a fundamental legal mismatch between the practices of training generative AI systems and the existing Text and Data Mining (TDM) exceptions outlined in the EU’s Copyright in the Digital Single Market (CDSM) Directive. These exceptions, particularly Article 4, were not designed to accommodate the expressive and synthetic nature of generative AI training. The study argues that applying these rules to AI risks distorting their original purpose and undermining the continued protection and fair remuneration of authors across Europe. The report outlines five key findings: (1) current EU TDM exceptions are inadequate for generative AI training; (2) fully machine-generated outputs should remain unprotected, while AI-assisted works need harmonized criteria; (3) a statutory remuneration scheme is crucial to bridge the value gap between creators and AI developers; (4) the fragmented governance landscape necessitates more coherent, cross-sector institutional responses; and (5) without timely reform, the EU faces legal uncertainty, market concentration, and cultural homogenization.
Input Side: The Debate Over Training Data
A central point of contention revolves around the use of copyright-protected content to train AI models. Generative AI systems, such as ChatGPT or Midjourney, require vast datasets, often scraped from the internet, which include copyrighted material. While Article 4 of the CDSM Directive allows TDM for any purpose, including commercial, unless rightsholders have explicitly ‘opted out’ using machine-readable means, this mechanism is widely criticized as ineffective. The report highlights that the ‘opt-out’ system is structurally unfit for large-scale AI training, lacking harmonized standards and imposing a disproportionate burden on individual creators. Many stakeholders, including the composer and songwriters lobby ECSA, argue that the TDM exceptions should not apply to generative AI and call for fair remuneration. Helienne Lindvall, president of ECSA, stated that MEP Axel Voss’s draft report “rejects the application of the TDM exceptions to Generative AI, and calls to ensure fair remuneration.” Conversely, the tech lobby CCIA maintains that the TDM exception is essential for AI innovation, with CCIA Europe’s Senior Policy Manager, Boniface de Champris, noting the rules were “carefully designed to strike a vital balance between fostering innovation and protecting intellectual property.”
Output Side: Human Authorship Remains Key
Regarding AI-generated content, the EU’s copyright law remains firmly human-centric. Works generated entirely by machines without human intervention do not qualify for copyright protection, as they lack the ‘author’s own intellectual creation’ required by EU jurisprudence. This means such outputs fall into the public domain. The report distinguishes between ‘AI-assisted’ works, where humans maintain creative control and make substantive choices, and ‘fully AI-generated’ outputs, which are not protectable. Merely providing a prompt to an AI model is generally not considered sufficient to establish authorship. This stance aims to preserve the public domain and prevent the monopolization of machine-generated content by platform owners.
Addressing the Value Gap and Remuneration
A significant concern is the absence of mechanisms to compensate creators whose works are used to train commercially valuable AI models. This ‘value gap’ undermines the incentive structure of copyright. The report explores various solutions, including statutory remuneration schemes, collective licensing, and revenue-sharing models. It draws inspiration from existing EU mechanisms like cable retransmission rights and the artist’s resale right, proposing a new EU-level statutory exception for AI training coupled with an unwaivable right to equitable remuneration, administered by collective management organizations (CMOs). This approach aims to provide a scalable and fair solution, acknowledging that individual licensing for billions of works is impractical.
AI Act’s Role and Limitations
The EU’s Artificial Intelligence Act (AI Act) introduces transparency obligations, requiring providers of general-purpose AI models to publish ‘sufficiently detailed summaries’ of their training data. However, many creators and industry groups, including GESAC, express skepticism. Burak Özgen, deputy general manager of GESAC, told Euractiv that the Commission’s template “falls short of safeguarding the creative sector and, if not corrected, risks undermining Europe’s AI Act and copyright framework in favour of a few global tech companies.” Critics argue that these summaries are often too vague to enable effective copyright enforcement and that the AI Act’s procedural nature does not resolve the underlying substantive copyright issues. The report emphasizes that transparency alone cannot substitute for a robust remuneration framework.
Path Forward: Policy Recommendations
Also Read:
- Animation Industry Grapples with Generative AI: A Divisive Tool Amidst Production Shifts, Luminate Report Reveals
- The Atlantic Unveils Tool for Creators to Monitor YouTube Content in AI Training Datasets
The parliamentary study proposes a multi-faceted approach for future reforms, focusing on legal clarity, transparency, and fair remuneration. Key recommendations include: establishing a permanent cross-sectoral governance platform and a JURI Working Group on AI and Copyright; launching a High-Level Expert Group to develop technical standards for opt-outs and remuneration; clarifying that Article 4 of the CDSM Directive does not extend to generative AI training, and instead advocating for an ‘opt-in’ principle for AI training; and developing a statutory remuneration mechanism for creators. These measures aim to ensure that AI development respects authors’ rights, prevents systemic imbalances, and promotes sustainable innovation in Europe’s creative ecosystem.
